Research On Virtual Optical Network Optimization Based On Comprehensive Quality Evaluation

As the scale of optical networks continues to grow,the existing network architecture becomes increasingly complex,and the difficulty of network management and operation increases.Networks are becoming more rigid,which has led to the emergence of network virtualization technology.By abstracting resources and dynamically loading them,network virtualization can distinguish user demands as independent virtual network requests and load them onto a common physical network,effectively solving the problem of network rigidity.With its flexible spectrum resource allocation,elastic optical networks have become an important network for carrying virtual optical networks and are gradually being valued and applied.After network virtualization is implemented,an important issue to consider is the resource mapping and allocation problem.Many new virtual optical network mapping algorithms have been proposed based on different optimization objectives.However,for network operators,the performance indicators for virtual optical network mapping are multidimensional,and algorithm performance needs to be considered from various aspects such as carrying capacity and survivability.Many factors in the mapping process can affect performance in these aspects,such as weight allocation,protection mode,and spectrum allocation method during the collaborative mapping process.Many existing studies have fixed mapping processes and a single evaluation index,making it a practical problem to comprehensively evaluate and select suitable mapping algorithms,which is the main research direction of this paper.This paper first studies the relevant indicators for evaluating the overall performance of virtual optical network mapping and selects classic indicators from the aspects of carrying capacity,reliability,survivability,and cost-benefit to construct a comprehensive evaluation index system for virtual optical networks.It can more comprehensively measure the performance of mapping algorithms.Next,the collaborative mapping algorithm is studied,the basic mapping framework is extracted,and the weight allocation,protection mode,and spectrum allocation algorithm modules in the process are decoupled.A multidimensional mapping algorithm loading model is established,which can more flexibly load different mapping algorithms and reasonably combine various modules in the algorithm process,forming the basis for the multi-algorithm comparison in this paper.Through simulation experiments and analysis,the effectiveness of different collaborative mapping algorithms and protection algorithms based on protection level are verified,and the performance differences of different algorithms in multi-dimensional indicators are demonstrated,which leads to the necessity of comprehensive quality evaluation.Regarding the issue of performance comparison of different algorithms,this paper studies the comprehensive quality evaluation model based on the improved EWM-TOPSIS.Different algorithms have their strengths and weaknesses in different indicators,making it difficult to directly evaluate the overall quality of the algorithms.Therefore,more accurate performance data is needed to be used for comprehensive evaluation based on evaluation methods.The entropy weighting method is used to determine the indicator weights after data preprocessing combined with the TOPSIS method to give the algorithm a comprehensive score for differentiation of the algorithms,providing a reference for algorithm selection.In practice,the network load is dynamically fluctuating,and it is found that different mapping algorithms perform differently under different loads in the mapping algorithm research.Therefore,this paper studies the dynamic algorithm allocation model based on load prediction.The LSTM model is used to predict the network load,and the average load within the upcoming business service time window is dynamically combined and scored to obtain a suitable mapping algorithm for mapping business requests.It is verified that this model can find the algorithm with better overall performance among different combinations of algorithms and ultimately optimize the overall mapping performance of virtual optical networks.